FINAL WEAKENING AND RELATED PHENOMENA1 Hans Henrich Hock University of Illinois at Urbana-Champaign 1: Final devoicing (FD) 1.1. In generative phonology, it is a generally accepted doctrine that, since word-final devoicing (WFD) is a very common and natural phenomenon, the ob- verse phenomenon, namely word-final voicin~ should not be found in natural language. Compare for instance Postal 1968 184 ('in the context----'~ the rules always devoice rather than voice'), Stampe 1969 443-5 (final devoicing comes about as the result of a failure t~ suppress the (innate) process of final devoicing), Vennemann 1972 240-1 (final voicing, defined as a process increasing the complexity of affected segment~ 'does not occur.')o 1.2 One of the standard examples for WFD is that of German, cf. Bund Bunde [bUnt] [bUndeJ. However Vennemann (1'9'6'8'""159-83 and in later publica- tions) and, following him, Hooper (1972 539) and Hyman (1975 142) have convincingly demonstrated that in Ger- man, this process applies not only word-finally, but also syllable-finally, as in radle [ra·t$le]3 'go by bike' (in some varieties of German). The standard view thus must be modified so as to recognize at least one other process, namely syllable-final devoicing (SFD). (For a different eA""Planation of this phenomenon compare section 2.3 below.) 2· Final voicing (or tenseness neutralization) 2 1 A more important argument against the stan- dard view, however, is that, as anyone with any train- ing in Indo-European linguistics can readily tell, there is at least one 5roup, namely Italic, where there is evidence for the allegedly impossible final voicing, cf PIE *siyet > OLat. sied 'would be'. 219 220 1975 MID-AMERICA LINGUISTICS CONFERENCE 2 2 According to Miller (1973 692), this is an example of 'an 11 unnatural 11 or 11 less natural" rule', FD being more natural--although the most 'natural thing is for nothing to happen in final position (as in English, French, etc.). There is no reason for a boundary to condition either voicing or devoicing The presence of either of these rules will be motivated by language-particular pause and boundary conditions.' At pauses, of course, 'the vocal cords must be open which vrill predict devoicing, if pause is anticipated ' ( fn 2, p 711) • Though Miller is, of course, correct in stating that nothing may happen at all, this is not a cogent argument against the claim that FD (or final voicing) may be a natural development For this claim evident- ly addresses its elf merely to the question '\vhat happens most naturally word-finally (or syllable-final- ly), if there is any change at all?' On the other hand, Inller' s argument that word ooundary is not a plausible phonetic environment for final (de)voic1ng is well taken, as is his claim thao pre~ position is a plausible phonetic environment, namely for (anticipatory) devoicing ( assimilatJon). As I have argued elsewhere (Hock, In press §5.1 w. fn 14), word-final devoicing, then, is to be explained as the result of (rule) generalization. However, this still \lould not account for final voicing 2 3 To my knowledge, it was Andersen (1969a,b) who first proposed an explanation of final voicing (not in Italic, buo in early Slavic) within the generative phonological paradigm. According to Andersen, the change of the preposition *ot to *od in dialectal Com- mon Slavic was the result, not of the usually recogrnzed process of final voice neutralization, but of final tenseness neutralization Just as in final voice neu- tralization it is the unmarked value of the feature voice], namely [- voice], which appears in the un- marked word-final position, so in final tenseness neu- tralizavion, the unmarked value of [+ appears, namely [-tense]. And it was the feature tense] FINAL WEAKENING AND RELATED PHENOMENA 2~1 which was 'phonemic' in early Slavic This 1s shown, in Andersen's vieu, by the fact that there was lenition (of~>£, > dz>!) in early (dialectal) Slavic. This explanation, though effective for early Slavic *at > *ad, hmJever, evidently causes problems in the case of-a language like German where [+ tense] is 'pho- nemic 1, 4 but where it is [+ tense] which appears in the word- or syllable-final position of neutralization (cf 1.2 above). To account for this phenomenon, Jmdersen (1972 44-5) assumes that unlike word-final position, the syllable-final position is 'marked', and that in this 'marked' environment it is the marked value of the feature [±tense], i.e. [+ tense], \Jhich appears. 2.4: This explanation of German SFD was accepted also by Anttila (1972:199-200) who argues that the syllable learning process in child language acquisrtion (first CV, then CVC, then VC, and finally V) shows that syllable-final environment i5 indeed 'marked in respect to syllable-initial position.' Moreover, he argues that 'Voicing i5 learned earlier in syllable-initial position, for example, \n1en an English-speaking child can already produce [buk] 'book' he goes on saying [pik] I pig I • • • I 2.5 A further boost for the hypo~hesis that 11 11 final voicing , i.e final laxing may be a genuine 1 alternative to FD came with Szemerenyi's paper of 1973. Szemerenyi first established that ~he final voi- cing of Italic has direct counterparts in Germanic and Celtic. I1oreover, he showed that there are early Iranian and Slavic counterparts to the rather unusual external voicing sandhi of Sanskrit [+stop]--> [c:it.voice] I __ # [~voice] with voicing sandhi across word boundary applying both before [M voice] and before [U voice] segments, while internally, the more natural assimilation [ + stop J --> [ voice] I [+ oostr. J °' -- °'voice is found, with voicing sandhi applying only before [r1 voice] segments. Thus, externally we find Skt 222 1975 MID-AMERICA LINGUISTICS CONFERENCE /vak#asya/ realized as vagasya 'his speech' vs. intern- al /vak-as/ --> vacas 'of speech'. It is only before obstruent that external and internal sandhi agree, as in /vak-bhis/ --> vagbhis 'by speeches' = /vak#dhiyate/ --> vagdhiyate 'speech is meditated (upon)'. Apparently without being aware of Andersen's (1969 a,b) similar argument for early Slavic, Szemerenyi then proposed to explain both of these sets of 'unusual' phenomena by assuming that in Proto-Indo-European, fin- al obstruents were neutralized in terms of the feature [±tense], in favor of the unmarked value, [-tense]. Italic, Celtic, and Germanic would then retain the resulting final [- tense] obstruents, while in Sanskrit, early Iranian, and early Slavic there must have been a generalization of the internal sandhi rule [+ stop] --> [~tense] I c: tobstr.] -- 111o ense to external sandmconditions. Note that as a conse- quence of this analysis, Sanskrit winds up \Ti th a more natural and 'usual' 'voicing assimilation' in external sandhi than under the traditional analysis in terms of the feature [±voice]. , 2.6 Though synchronically, Andersen's and Szeme- renyi's analysis may well be correct, I am not con- vinced that it must be correct also historically 2.6.1 First of all, as I hope to have shown else- where (Hock, In press.~5.1), it is possible to account for the external voicing sand.hi of Sanskrit (as well as of early Iranian and Slavic) as the result of an ana- logical generalization of final devoicing, reinterpreted as voice neutralization. Such a neutralization can then oe manif estea through the rule [+ stop] --> [~ voice] I # [OC voice] by overgeneralization of the internal voicing rule [+ stop] --> [«voice] I [+ obstr.] -- OC. voice • 'As a generalization (of voice neuLJralization before all(') instances of word boundary), this process is not bound by the same conditions of phonetic natural- FINAL WEAKENING AND RELATED PHENOMENA 223 ness as a purely phonetically conditioned change.' There is ample evidence supporting this assumption, both in terms of other sandhi (over)general1zat1ons in Sanskrit, and in terms of parallel developments in Polish dialects (cf Vaillant 1950 101 and 209) It is interesting to note that Andersen (1969b 302-3 w. earlier literature) provides conclusive evid- ence for a similar generalization in Slovak, as in (*su vod(o~)9 >) s vodou [z vodou] 'with water'. Relics of an earlier stage, in which there was no such over- generalization of voice assimilation are found in ex- pressions containing personal or anaphoric pronouns, as ins nam1 [s nami] 'with us'. It is then possible to argue that this overgener- alization may have occurred in Froto-Indo-European and that, because of the resulting occurrence of voiced stops in most environments, it was possible to take these voiced stops as the basic word-final manifesta- tion of stops, and- to generalize~ them, presumably re- interpreted as [- tense] stops, to all environments. This uould account for the Italic, Celtic, and Germanic situation .. Note that this explanation thus removes the dif- ficulties of Brugmann (1897 883-4, 885-6) who uas in- clined to view the Sanskrit external sandhi pattern as analogical,) but felt that in light of Italic *-t > -d, it may have been inheritedfrom Proto-Indo-European-;-6 thus presumably precluding an analogical origin. Moreover, the analysis here proposed as a possible alternative has the advantage that, unlike that of An- dersen and Szemerenyi, it offers a phonetically motiv- ated starting point for the phenomena under discussion~ namely (prepausal) FD. (But cf also section 3 below.; Finally, it might be argued that the evidence of the extant Sanskrit texts, where the prepausal form of final obstruents invariably is voiceless (as in /pad##/ --> p@t 'foot') would be more compatible with an analy- sis w ich postulates voicing, rather than tenseness neutralization. However, as the subsequent discussion 224 1975 MID-AMERICA LINGUISTICS CONFERENCE in this paper will show, the Sanskrit facts are open to various interpretations.